Ultrafast non-reciprocal phase transitions in Kagome spin ice plasmonic metasurfaces

We present a high-sensitivity method to investigate the ultrafast spin dynamics of artificial spin ice (ASI) systems. ASI metasurfaces, comprising coupled nanomagnets in frustrated lattices, offer immense potential for low-energy information processing through non-reciprocal information flow, which is crucial for next generation in-memory microelectronic computing architectures like probabilistic and neuromorphic computing. Using magneto-optic Kerr effect (MOKE) measurements at the plasmonic wavelength (900 nm), we reveal non-reciprocal spin resonances in both longitudinal (L-MOKE) and polar (P-MOKE) configurations. We demonstrate the efficacy of the metal-insulator-metal plasmonic resonance, which amplifies the MOKE response by two orders of magnitude allowing us to achieve unparalleled sensitivity in detecting magnetic phase transitions in the Kagome nanomagnetic lattice matching micromagnetic simulations and magnetic force microscopy measurements. Our findings demonstrate the potential for all-optical reconfiguration and readout in spin ice systems, laying the groundwork for applications in stochastic learning architectures such as directional graph networks and deep reservoirs.